HELM VHELM - Blink Perception - Relative Depth: leaderboard
VHELM (Stanford CRFM HELM for vision-language models, run v2.1.2) BLINK Relative Depth task: which of two marked points is closer to the camera, one of BLINK's perception tasks that humans solve at a glance, asked as multiple choice; 124 questions, prefix quasi-exact match. Percent correct, 0-100, higher is better.
Metric: Prefix Quasi-Exact Match (%). Source: crfm.stanford.edu. Saturation forecast: Estimated already saturated. 48 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 4 Scout Instruct | 96.77 | #547 |
| 2 | Llama 4 Maverick Instruct FP8 | 91.94 | #424 |
| 3 | Gemini 2.5 Pro (Preview 03-25) | 90.32 | #126 |
| 4 | Claude 3 Opus (20240229) | 89.52 | #428 |
| 5 | GPT-4.5 | 88.71 | #278 |
| 6 | GPT-4.1 (2025-04-14) | 87.1 | #257 |
| 7 | Gemini 2.0 Flash Lite | 87.1 | #455 |
| 8 | O3 (2025-04-16) | 86.29 | #125 |
| 9 | Gemini 1.5 Pro (002) | 84.68 | #368 |
| 10 | Gemini 1.5 Flash (002) | 83.87 | #587 |
| 11 | Gemini 2.0 Pro (Preview 02-05) | 83.87 | #318 |
| 12 | Gemini 2.0 Flash | 83.06 | #351 |
| 13 | O4 Mini (2025-04-16) | 83.06 | #183 |
| 14 | Claude 3 Sonnet (20240229) | 82.26 | #694 |
| 15 | GPT-4.1 Mini | 81.45 | #363 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-vhelm-blink-perception-relative-depth · How It Works · Data refreshed daily, snapshot 2026-11-05.